Systems and Methods for Privacy Preserving Fraud Detection During Electronic Transactions

1. A method for performing privacy-preserving fraud detection in network-based network operations, the method comprising: receiving, by a server system from a user computing system, a fraud detection message during a network operation between the user computing system and a merchant computing system, the fraud detection message having a set of cryptographically transformed universal resource locator (URL) components generated from a URL of a web page of the merchant computing system on which the network operation is to occur; parsing, by the server system, the set of cryptographically transformed URL components of the web page to separate each cryptographically transformed URL component; generating, by the server system, one or more secure and anonymous fraud detection features, each fraud detection feature comprising a select subset of the separated cryptographically transformed URL components; executing, by the server system, a machine learning model that uses the one or more secure and anonymous fraud detection features to determine a likelihood that fraud is occurring in the network operation; determining, by the server system, that the likelihood of fraud occurring in the network operation satisfies a fraud detection threshold; incrementing, by the server system, a count of detected fraud for at least one of the one or more secure and anonymous fraud detection features to generate a fraud count associated with the at least one of the one or more secure and anonymous fraud detection features; and training, by the server system, the machine learning model based at least in part on the incremented fraud count associated with the at least one of the one or more secure and anonymous fraud detection features.

2. The method of claim 1, wherein performing fraud detection for the web page using the one or more secure and anonymous fraud detection features comprises: determining, by the server system, for each of the one or more secure and anonymous fraud detection features, a corresponding fraud count indicative of past detected fraud associated with said each of the one or more secure and anonymous fraud detection features; processing, by the server system, the one or more secure and anonymous fraud detection features and corresponding fraud counts by the machine learning model trained to detect fraud based on cryptographically transformed URL components and associated fraud counts; and generating, by the machine learning model, a prediction for the likelihood that fraud is occurring in the network operation.

3. The method of claim 2, further comprising: when the prediction for the likelihood that fraud is occurring in the network operation satisfies the fraud detection threshold, incrementing, by the server system, the count of detected fraud for each of the one or more secure and anonymous fraud detection features to generate an updated fraud count associated with each of the one or more secure and anonymous fraud detection features; and performing retraining, by the server system, of the machine learning model based on one or more updated fraud counts comprising the updated fraud count.

4. The method of claim 1, further comprising: in response to a determination that the likelihood that fraud is occurring in the network operation satisfies a fraud detection threshold, performing, by the server system, one or more fraud remediation processes comprising declining the network operation between the user computing system and a merchant system, transmitting a transaction declined message for rendering in the web page at the merchant system, and generating a notification to the merchant system of the detected fraudulent network operation.

5. The method of claim 4, in response to a determination that the likelihood that fraud is occurring in the network operation does not satisfy the fraud detection threshold, completing, by the server system, the network operation between the user computing system and the merchant system.

6. The method of claim 1, wherein the set of cryptographically transformed universal resource locator (URL) components are formatted into a fraud detection URL, and each cryptographically transformed URL component in the fraud detection URL corresponds to an original URL component from the web page.

7. The method of claim 6, wherein each cryptographically transformed URL component is a hash value generated using the corresponding original URL component from the web page.

8. The method of claim 6, wherein parsing the set of cryptographically transformed URL components of the web page to separate each cryptographically transformed URL component, comprises: detecting, by the server system, one or more special characters within the fraud detection URL that delineate individual cryptographically transformed universal resource locator (URL) components within the fraud detection URL; and extracting, by the server system, each of the individual cryptographically transformed universal resource locator (URL) components, as delineated by the detected one or more special characters, from the fraud detection URL.

9. The method of claim 8, wherein each fraud detection feature, comprising the select subset of the separated each cryptographically transformed URL components, comprises at least one cryptographically transformed URL component, and said each fraud detection feature further comprises data indicative of a function of the at least one cryptographically transformed URL component determined based on detected one or more special characters used to extract the at least one cryptographically transformed URL component from the fraud detection URL.

10. The method of claim 1, wherein the fraud detection message comprises a fraud detection payload that comprises the set of cryptographically transformed universal resource locator (URL) components.

11. The method of claim 1, wherein the cryptographically transformed URL components are generated by server system web page components executing within the web page, the server system web page components comprising one or more functions, API(s), SDK(s), or a combination thereof, distributed by a commerce platform and integrated into the web page by a merchant system prior to the web page being served to the user computing system during the network operation.

12. The method of claim 11, further comprising: accessing, by the server system web page components executing within the web page, the URL of the web page; parsing, by the server system web page components, the URL of the web page into component parts of the web page based on special characters that delineate each component part within the URL of the web page; cryptographically transforming, by the server system web page components, each component part of the web page; assembling, by the server system web page components, the cryptographically transformed component parts of the web page into a fraud detection URL having a format of the URL of the web page, the cryptographically transformed component parts within the fraud detection URL comprising the set of cryptographically transformed universal resource locator (URL) components; adding, by the server system web page components, the fraud detection URL to a payload of the fraud detection message; and transmitting, from the server system web page components to a web server, the fraud detection message.

13. The method of claim 12, wherein the assembling further comprises: compressing, by the server system, the assembled set of cryptographically transformed universal resource locator (URL) components by truncating at least one of the cryptographically transformed universal resource locator (URL) components, combining two or more component parts of the web page and cryptographically transforming the combined two or more component parts to generate a single cryptographically transformed component part of the web page, or a combination thereof; and assembling, by the server system, the fraud detection URL after the compressing.

14. The method of claim 13, wherein the truncating is performed for one or more cryptographically transformed universal resource locator (URL) components that exceed one of a plurality of thresholds, each threshold being associated with a type of the one or more cryptographically transformed universal resource locator (URL) components.

15. The method of claim 13, further comprising: combining, by the server system, the two or more component parts of the web page when a total number of component parts of the URL of the web page exceed a maximum number of components parts, and wherein the two or more component parts of the web page being combined comprise two or more component parts located at an end of the URL of the web page that exceed the maximum number of components parts.

16. The method of claim 1, wherein the server system is a commerce platform system performing one or more payment processing services for a merchant system during the network operation.

17. A non-transitory computer-readable storage medium including instructions that, when executed by a processor, cause the processor to perform operations for performing privacy preserving fraud detection in network-based network operations, the operations comprising: receiving, by a server system from a user system, a fraud detection message during a network operation between the user system and a merchant system, the fraud detection message having a set of cryptographically transformed universal resource locator (URL) components generated from a URL of a web page of the merchant system on which the network operation is to occur; parsing, by the server system, the set of cryptographically transformed URL components of the web page to separate each cryptographically transformed URL component; generating, by the server system, one or more secure and anonymous fraud detection features, each fraud detection feature comprising a select subset of the separated each cryptographically transformed URL components; executing, by the server system, a machine learning model that uses the one or more secure and anonymous fraud detection features to determine a likelihood that fraud is occurring in the network operation; determine that the likelihood of fraud occurring in the network operation satisfies a fraud detection threshold; increment a count of detected fraud for at least one of the one or more secure and anonymous fraud detection features to generate a fraud count associated with the at least one of the one or more secure and anonymous fraud detection features; and train the machine learning model based at least in part on the incremented fraud count associated with the at least one of the one or more secure and anonymous fraud detection features.

18. The non-transitory computer-readable storage medium of claim 17, wherein performing fraud detection for the web page using the one or more secure and anonymous fraud detection features comprises: determining, for each of the one or more secure and anonymous fraud detection features, a corresponding fraud count indicative of past detected fraud associated with said each of the one or more secure and anonymous fraud detection features; processing the one or more secure and anonymous fraud detection features and corresponding fraud counts by a machine learning model trained to detect fraud based on cryptographically transformed URL components and associated fraud counts; and generating, by the machine learning model, a prediction for the likelihood that fraud is occurring in the network operation.

19. The non-transitory computer-readable storage medium of claim 17, wherein the set of cryptographically transformed universal resource locator (URL) components are formatted into a fraud detection URL, and each cryptographically transformed URL component in the fraud detection URL corresponds to an original URL component from the web page.

20. The non-transitory computer-readable storage medium of claim 19, wherein each cryptographically transformed URL component is a hash value generated using the corresponding original URL component from the web page.

21. The non-transitory computer-readable storage medium of claim 17, wherein the cryptographically transformed URL components are generated by server system web page components executing within the web page, the server system web page components comprising one or more functions, API(s), SDK(s), or a combination thereof, distributed by a commerce platform and integrated into the web page by the merchant system prior to the web page being served to the user system during the network operation.

22. A server computer system for performing privacy-preserving fraud detection in network-based network operation, comprising: a memory; and a processor coupled with the memory configured to: receive, from a user system, a fraud detection message during a network operation between the user system and a merchant system, the fraud detection message having a set of cryptographically transformed universal resource locator (URL) components generated from a URL of a web page of the merchant system on which the network operation is to occur, parse the set of cryptographically transformed URL components of the web page to separate each cryptographically transformed URL component, generate one or more secure and anonymous fraud detection features, each fraud detection feature comprising a select subset of the separated each cryptographically transformed URL components, execute a machine learning model that uses the one or more secure and anonymous fraud detection features to determine a likelihood that fraud is occurring in the network operation; determine that the likelihood of fraud occurring in the network operation satisfies a fraud detection threshold; increment a count of detected fraud for at least one of the one or more secure and anonymous fraud detection features to generate a fraud count associated with the at least one of the one or more secure and anonymous fraud detection features; and train the machine learning model based at least in part on the incremented fraud count associated with the at least one of the one or more secure and anonymous fraud detection features.